Translated texts exhibit systematic linguistic differences compared to original texts in the same language, and these differences are referred to as translationese. Translationese has effects on various cross-lingual natural language processing tasks, potentially leading to biased results. In this paper, we explore a novel approach to reduce translationese in translated texts: translation-based style transfer. As there are no parallel human-translated and original data in the same language, we use a self-supervised approach that can learn from comparable (rather than parallel) mono-lingual original and translated data. However, even this self-supervised approach requires some parallel data for validation. We show how we can eliminate the need for parallel validation data by combining the self-supervised loss with an unsupervised loss. This unsupervised loss leverages the original language model loss over the style-transferred output and a semantic similarity loss between the input and style-transferred output. We evaluate our approach in terms of original vs. translationese binary classification in addition to measuring content preservation and target-style fluency. The results show that our approach is able to reduce translationese classifier accuracy to a level of a random classifier after style transfer while adequately preserving the content and fluency in the target original style.

Recent work has shown evidence of “Clever Hans” behavior in high-performance neural translationese classifiers, where BERT-based classifiers capitalize on spurious correlations, in particular topic information, between data and target classification labels, rather than genuine translationese signals. Translationese signals are subtle (especially for professional translation) and compete with many other signals in the data such as genre, style, author, and, in particular, topic. This raises the general question of how much of the performance of a classifier is really due to spurious correlations in the data versus the signals actually targeted for by the classifier, especially for subtle target signals and in challenging (low resource) data settings. We focus on topic-based spurious correlation and approach the question from two directions: (i) where we have no knowledge about spurious topic information and its distribution in the data, (ii) where we have some indication about the nature of spurious topic correlations. For (i) we develop a measure from first principles capturing alignment of unsupervised topics with target classification labels as an indication of spurious topic information in the data. We show that our measure is the same as purity in clustering and propose a “topic floor” (as in a “noise floor”) for classification. For (ii) we investigate masking of known spurious topic carriers in classification. Both (i) and (ii) contribute to quantifying and (ii) to mitigating spurious correlations.

This paper presents the results of the Second WMT Shared Task on Sign Language Translation (WMT-SLT23; https://www.wmt-slt.com/). This shared task is concerned with automatic translation between signed and spoken languages. The task is unusual in the sense that it requires processing visual information (such as video frames or human pose estimation) beyond the well-known paradigm of text-to-text machine translation (MT). The task offers four tracks involving the following languages: Swiss German Sign Language (DSGS), French Sign Language of Switzerland (LSF-CH), Italian Sign Language of Switzerland (LIS-CH), German, French and Italian. Four teams (including one working on a baseline submission) participated in this second edition of the task, all submitting to the DSGS-to-German track. Besides a system ranking and system papers describing state-of-the-art techniques, this shared task makes the following scientific contributions: novel corpora and reproducible baseline systems. Finally, the task also resulted in publicly available sets of system outputs and more human evaluation scores for sign language translation.

This paper is a brief summary of the First WMT Shared Task on Sign Language Translation (WMT-SLT22), a project partly funded by EAMT. The focus of this shared task is automatic translation between signed and spoken languages. Details can be found on our website (https://www.wmt-slt.com/) or in the findings paper (Müller et al., 2022).

We present the first neural machine translation system for the low-resource language pair Wayúunaiki–Spanish and explore strategies to inject linguistic knowledge into the model to improve translation quality. We explore a wide range of methods and combine complementary approaches. Results indicate that incorporating linguistic information through linguistically motivated subword segmentation, factored models, and pretrained embeddings helps the system to generate improved translations, with the segmentation contributing most. In order to evaluate translation quality in a general domain and go beyond the available religious domain data, we gather and make publicly available a new test set and supplementary material. Although translation quality as measured with automatic metrics is low, we hope these resources will facilitate and support further research on Wayúunaiki.

Neural language models play an increasingly central role for language processing, given their success for a range of NLP tasks. In this study, we compare some canonical strategies in language modeling for low-resource scenarios, evaluating all models by their (finetuned) performance on a POS-tagging downstream task. We work with five (extremely) low-resource dialects from the Indic dialect continuum (Braj, Awadhi, Bhojpuri, Magahi, Maithili), which are closely related to each other and the standard mid-resource dialect, Hindi. The strategies we evaluate broadly include from-scratch pretraining, and cross-lingual transfer between the dialects as well as from different kinds of off-the- shelf multilingual models; we find that a model pretrained on other mid-resource Indic dialects and languages, with extended pretraining on target dialect data, consistently outperforms other models. We interpret our results in terms of dataset sizes, phylogenetic relationships, and corpus statistics, as well as particularities of this linguistic system.

Dense vector representations for textual data are crucial in modern NLP. Word embeddings and sentence embeddings estimated from raw texts are key in achieving state-of-the-art resultsin various tasks requiring semantic understanding. However, obtaining embeddings at the document level is challenging due to computational requirements and lack of appropriate data. Instead, most approaches fall back on computing document embeddings based on sentence representations. Although there exist architectures and models to encode documents fully, they are in general limited to English and few other high-resourced languages. In this work, we provide a systematic comparison of methods to produce document-level representations from sentences based on LASER, LaBSE, and Sentence BERT pre-trained multilingual models. We compare input token number truncation, sentence averaging as well as some simple windowing and in some cases new augmented and learnable approaches, on 3 multi- and cross-lingual tasks in 8 languages belonging to 3 different language families. Our task-based extrinsic evaluations show that, independently of the language, a clever combination of sentence embeddings is usually better than encoding the full document as a single unit, even when this is possible. We demonstrate that while a simple sentence average results in a strong baseline for classification tasks, more complex combinations are necessary for semantic tasks

Neutrality is difficult to achieve and, in politics, subjective. Traditional media typically adopt an editorial line that can be used by their potential readers as an indicator of the media bias. Several platforms currently rate news outlets according to their political bias. The editorial line and the ratings help readers in gathering a balanced view of news. But in the advent of instruction-following language models, tasks such as writing a newspaper article can be delegated to computers. Without imposing a biased persona, where would an AI-based news outlet lie within the bias ratings? In this work, we use the ratings of authentic news outlets to create a multilingual corpus of news with coarse stance annotations (Left and Right) along with automatically extracted topic annotations. We show that classifiers trained on this data are able to identify the editorial line of most unseen newspapers in English, German, Spanish and Catalan. We then apply the classifiers to 101 newspaper-like articles written by ChatGPT and Bard in the 4 languages at different time periods. We observe that, similarly to traditional newspapers, ChatGPT editorial line evolves with time and, being a data-driven system, the stance of the generated articles differs among languages.

Some human preferences are universal. The odor of vanilla is perceived as pleasant all around the world. We expect neural models trained on human texts to exhibit these kind of preferences, i.e. biases, but we show that this is not always the case. We explore 16 static and contextual embedding models in 9 languages and, when possible, compare them under similar training conditions. We introduce and release CA-WEAT, multilingual cultural aware tests to quantify biases, and compare them to previous English-centric tests. Our experiments confirm that monolingual static embeddings do exhibit human biases, but values differ across languages, being far from universal. Biases are less evident in contextual models, to the point that the original human association might be reversed. Multilinguality proves to be another variable that attenuates and even reverses the effect of the bias, specially in contextual multilingual models. In order to explain this variance among models and languages, we examine the effect of asymmetries in the training corpus, departures from isomorphism in multilingual embedding spaces and discrepancies in the testing measures between languages.

Recent research on style transfer takes inspiration from unsupervised neural machine translation (UNMT), learning from large amounts of non-parallel data by exploiting cycle consistency loss, back-translation, and denoising autoencoders. By contrast, the use of selfsupervised NMT (SSNMT), which leverages (near) parallel instances hidden in non-parallel data more efficiently than UNMT, has not yet been explored for style transfer. In this paper we present a novel Self-Supervised Style Transfer (3ST) model, which augments SSNMT with UNMT methods in order to identify and efficiently exploit supervisory signals in non-parallel social media posts. We compare 3ST with state-of-the-art (SOTA) style transfer models across civil rephrasing, formality and polarity tasks. We show that 3ST is able to balance the three major objectives (fluency, content preservation, attribute transfer accuracy) the best, outperforming SOTA models on averaged performance across their tested tasks in automatic and human evaluation.

We present a novel method for unsupervised cognate/borrowing identification from monolingual corpora designed for low and extremely low resource scenarios, based on combining noisy semantic signals from joint bilingual spaces with orthographic cues modelling sound change. We apply our method to the North Indian dialect continuum, containing several dozens of dialects and languages spoken by more than 100 million people. Many of these languages are zero-resource and therefore natural language processing for them is non-existent. We first collect monolingual data for 26 Indic languages, 16 of which were previously zero-resource, and perform exploratory character, lexical and subword cross-lingual alignment experiments for the first time at this scale on this dialect continuum. We create bilingual evaluation lexicons against Hindi for 20 of the languages. We then apply our cognate identification method on the data, and show that our method outperforms both traditional orthography baselines as well as EM-style learnt edit distance matrices. To the best of our knowledge, this is the first work to combine traditional orthographic cues with noisy bilingual embeddings to tackle unsupervised cognate detection in a (truly) low-resource setup, showing that even noisy bilingual embeddings can act as good guides for this task. We release our multilingual dialect corpus, called HinDialect, as well as our scripts for evaluation data collection and cognate induction.

This paper presents the results of the First WMT Shared Task on Sign Language Translation (WMT-SLT22).This shared task is concerned with automatic translation between signed and spoken languages. The task is novel in the sense that it requires processing visual information (such as video frames or human pose estimation) beyond the well-known paradigm of text-to-text machine translation (MT).The task featured two tracks, translating from Swiss German Sign Language (DSGS) to German and vice versa. Seven teams participated in this first edition of the task, all submitting to the DSGS-to-German track. Besides a system ranking and system papers describing state-of-the-art techniques, this shared task makes the following scientific contributions: novel corpora, reproducible baseline systems and new protocols and software for human evaluation. Finally, the task also resulted in the first publicly available set of system outputs and human evaluation scores for sign language translation.

This paper describes the DFKI-MLT submission to the WMT-SLT 2022 sign language translation (SLT) task from Swiss German Sign Language (video) into German (text). State-of-the-art techniques for SLT use a generic seq2seq architecture with customized input embeddings. Instead of word embeddings as used in textual machine translation, SLT systems use features extracted from video frames. Standard approaches often do not benefit from temporal features. In our participation, we present a system that learns spatio-temporal feature representations and translation in a single model, resulting in a real end-to-end architecture expected to better generalize to new data sets. Our best system achieved 5 ± 1 BLEU points on the development set, but the performance on the test dropped to 0.11 ± 0.06 BLEU points.

Cross-lingual natural language processing relies on translation, either by humans or machines, at different levels, from translating training data to translating test sets. However, compared to original texts in the same language, translations possess distinct qualities referred to as translationese. Previous research has shown that these translation artifacts influence the performance of a variety of cross-lingual tasks. In this work, we propose a novel approach to reducing translationese by extending an established bias-removal technique. We use the Iterative Null-space Projection (INLP) algorithm, and show by measuring classification accuracy before and after debiasing, that translationese is reduced at both sentence and word level. We evaluate the utility of debiasing translationese on a natural language inference (NLI) task, and show that by reducing this bias, NLI accuracy improves. To the best of our knowledge, this is the first study to debias translationese as represented in latent embedding space.

Recent work has shown that neural feature- and representation-learning, e.g. BERT, achieves superior performance over traditional manual feature engineering based approaches, with e.g. SVMs, in translationese classification tasks. Previous research did not show (i) whether the difference is because of the features, the classifiers or both, and (ii) what the neural classifiers actually learn. To address (i), we carefully design experiments that swap features between BERT- and SVM-based classifiers. We show that an SVM fed with BERT representations performs at the level of the best BERT classifiers, while BERT learning and using handcrafted features performs at the level of an SVM using handcrafted features. This shows that the performance differences are due to the features. To address (ii) we use integrated gradients and find that (a) there is indication that information captured by hand-crafted features is only a subset of what BERT learns, and (b) part of BERT’s top performance results are due to BERT learning topic differences and spurious correlations with translationese.

Previous research has used linguistic features to show that translations exhibit traces of source language interference and that phylogenetic trees between languages can be reconstructed from the results of translations into the same language. Recent research has shown that instances of translationese (source language interference) can even be detected in embedding spaces, comparing embeddings spaces of original language data with embedding spaces resulting from translations into the same language, using a simple Eigenvector-based divergence from isomorphism measure. To date, it remains an open question whether alternative graph-isomorphism measures can produce better results. In this paper, we (i) explore Gromov-Hausdorff distance, (ii) present a novel spectral version of the Eigenvector-based method, and (iii) evaluate all approaches against a broad linguistic typological database (URIEL). We show that language distances resulting from our spectral isomorphism approaches can reproduce genetic trees on a par with previous work without requiring any explicit linguistic information and that the results can be extended to non-Indo-European languages. Finally, we show that the methods are robust under a variety of modeling conditions.

Massively multilingual machine translation (MT) has shown impressive capabilities and including zero and few-shot translation between low-resource language pairs. However and these models are often evaluated on high-resource languages with the assumption that they generalize to low-resource ones. The difficulty of evaluating MT models on low-resource pairs is often due to lack of standardized evaluation datasets. In this paper and we present MENYO-20k and the first multi-domain parallel corpus with a especially curated orthography for Yoruba–English with standardized train-test splits for benchmarking. We provide several neural MT benchmarks and compare them to the performance of popular pre-trained (massively multilingual) MT models both for the heterogeneous test set and its subdomains. Since these pre-trained models use huge amounts of data with uncertain quality and we also analyze the effect of diacritics and a major characteristic of Yoruba and in the training data. We investigate how and when this training condition affects the final quality of a translation and its understandability.Our models outperform massively multilingual models such as Google (+8.7 BLEU) and Facebook M2M (+9.1) when translating to Yoruba and setting a high quality benchmark for future research.

For most language combinations and parallel data is either scarce or simply unavailable. To address this and unsupervised machine translation (UMT) exploits large amounts of monolingual data by using synthetic data generation techniques such as back-translation and noising and while self-supervised NMT (SSNMT) identifies parallel sentences in smaller comparable data and trains on them. To this date and the inclusion of UMT data generation techniques in SSNMT has not been investigated. We show that including UMT techniques into SSNMT significantly outperforms SSNMT (up to +4.3 BLEU and af2en) as well as statistical (+50.8 BLEU) and hybrid UMT (+51.5 BLEU) baselines on related and distantly-related and unrelated language pairs.

This paper presents an overview of AVASAG; an ongoing applied-research project developing a text-to-sign-language translation system for public services. We describe the scientific innovation points (geometry-based SL-description, 3D animation and video corpus, simplified annotation scheme, motion capture strategy) and the overall translation pipeline.

Traditional hand-crafted linguistically-informed features have often been used for distinguishing between translated and original non-translated texts. By contrast, to date, neural architectures without manual feature engineering have been less explored for this task. In this work, we (i) compare the traditional feature-engineering-based approach to the feature-learning-based one and (ii) analyse the neural architectures in order to investigate how well the hand-crafted features explain the variance in the neural models’ predictions. We use pre-trained neural word embeddings, as well as several end-to-end neural architectures in both monolingual and multilingual settings and compare them to feature-engineering-based SVM classifiers. We show that (i) neural architectures outperform other approaches by more than 20 accuracy points, with the BERT-based model performing the best in both the monolingual and multilingual settings; (ii) while many individual hand-crafted translationese features correlate with neural model predictions, feature importance analysis shows that the most important features for neural and classical architectures differ; and (iii) our multilingual experiments provide empirical evidence for translationese universals across languages.

This paper presents the results of the newstranslation task, the multilingual low-resourcetranslation for Indo-European languages, thetriangular translation task, and the automaticpost-editing task organised as part of the Con-ference on Machine Translation (WMT) 2021.In the news task, participants were asked tobuild machine translation systems for any of10 language pairs, to be evaluated on test setsconsisting mainly of news stories. The taskwas also opened up to additional test suites toprobe specific aspects of translation.

Self-supervised neural machine translation (SSNMT) jointly learns to identify and select suitable training data from comparable (rather than parallel) corpora and to translate, in a way that the two tasks support each other in a virtuous circle. In this study, we provide an in-depth analysis of the sampling choices the SSNMT model makes during training. We show how, without it having been told to do so, the model self-selects samples of increasing (i) complexity and (ii) task-relevance in combination with (iii) performing a denoising curriculum. We observe that the dynamics of the mutual-supervision signals of both system internal representation types are vital for the extraction and translation performance. We show that in terms of the Gunning-Fog Readability index, SSNMT starts extracting and learning from Wikipedia data suitable for high school students and quickly moves towards content suitable for first year undergraduate students.

Translationese is a phenomenon present in human translations, simultaneous interpreting, and even machine translations. Some translationese features tend to appear in simultaneous interpreting with higher frequency than in human text translation, but the reasons for this are unclear. This study analyzes translationese patterns in translation, interpreting, and machine translation outputs in order to explore possible reasons. In our analysis we – (i) detail two non-invasive ways of detecting translationese and (ii) compare translationese across human and machine translations from text and speech. We find that machine translation shows traces of translationese, but does not reproduce the patterns found in human translation, offering support to the hypothesis that such patterns are due to the model (human vs machine) rather than to the data (written vs spoken).

The success of several architectures to learn semantic representations from unannotated text and the availability of these kind of texts in online multilingual resources such as Wikipedia has facilitated the massive and automatic creation of resources for multiple languages. The evaluation of such resources is usually done for the high-resourced languages, where one has a smorgasbord of tasks and test sets to evaluate on. For low-resourced languages, the evaluation is more difficult and normally ignored, with the hope that the impressive capability of deep learning architectures to learn (multilingual) representations in the high-resourced setting holds in the low-resourced setting too. In this paper we focus on two African languages, Yorùbá and Twi, and compare the word embeddings obtained in this way, with word embeddings obtained from curated corpora and a language-dependent processing. We analyse the noise in the publicly available corpora, collect high quality and noisy data for the two languages and quantify the improvements that depend not only on the amount of data but on the quality too. We also use different architectures that learn word representations both from surface forms and characters to further exploit all the available information which showed to be important for these languages. For the evaluation, we manually translate the wordsim-353 word pairs dataset from English into Yorùbá and Twi. We extend the analysis to contextual word embeddings and evaluate multilingual BERT on a named entity recognition task. For this, we annotate with named entities the Global Voices corpus for Yorùbá. As output of the work, we provide corpora, embeddings and the test suits for both languages.

We introduce GeBioToolkit, a tool for extracting multilingual parallel corpora at sentence level, with document and gender information from Wikipedia biographies. Despite the gender inequalities present in Wikipedia, the toolkit has been designed to extract corpus balanced in gender. While our toolkit is customizable to any number of languages (and different domains), in this work we present a corpus of 2,000 sentences in English, Spanish and Catalan, which has been post-edited by native speakers to become a high-quality dataset for machine translation evaluation. While GeBioCorpus aims at being one of the first non-synthetic gender-balanced test datasets, GeBioToolkit aims at paving the path to standardize procedures to produce gender-balanced datasets.

We introduce the Computational Linguistics special issue on Multilingual and Interlingual Semantic Representations for Natural Language Processing. We situate the special issue’s five articles in the context of our fast-changing field, explaining our motivation for this project. We offer a brief summary of the work in the issue, which includes developments on lexical and sentential semantic representations, from symbolic and neural perspectives.

Recent studies use a combination of lexical and syntactic features to show that footprints of the source language remain visible in translations, to the extent that it is possible to predict the original source language from the translation. In this paper, we focus on embedding-based semantic spaces, exploiting departures from isomorphism between spaces built from original target language and translations into this target language to predict relations between languages in an unsupervised way. We use different views of the data — words, parts of speech, semantic tags and synsets — to track translationese. Our analysis shows that (i) semantic distances between original target language and translations into this target language can be detected using the notion of isomorphism, (ii) language family ties with characteristics similar to linguistically motivated phylogenetic trees can be inferred from the distances and (iii) with delexicalised embeddings exhibiting source-language interference most significantly, other levels of abstraction display the same tendency, indicating the lexicalised results to be not “just” due to possible topic differences between original and translated texts. To the best of our knowledge, this is the first time departures from isomorphism between embedding spaces are used to track translationese.

This paper describes the UdS-DFKI submission to the WMT2019 news translation task for Gujarati–English (low-resourced pair) and German–English (document-level evaluation). Our systems rely on the on-line extraction of parallel sentences from comparable corpora for the first scenario and on the inclusion of coreference-related information in the training data in the second one.

We analyse coreference phenomena in three neural machine translation systems trained with different data settings with or without access to explicit intra- and cross-sentential anaphoric information. We compare system performance on two different genres: news and TED talks. To do this, we manually annotate (the possibly incorrect) coreference chains in the MT outputs and evaluate the coreference chain translations. We define an error typology that aims to go further than pronoun translation adequacy and includes types such as incorrect word selection or missing words. The features of coreference chains in automatic translations are also compared to those of the source texts and human translations. The analysis shows stronger potential translationese effects in machine translated outputs than in human translations.

This work presents a decoding architecture that fuses the information from a neural translation model and the context semantics enclosed in a semantic space language model based on word embeddings. The method extends the beam search decoding process and therefore can be applied to any neural machine translation framework. With this, we sidestep two drawbacks of current document-level systems: (i) we do not modify the training process so there is no increment in training time, and (ii) we do not require document-level an-notated data. We analyze the impact of the fusion system approach and its parameters on the final translation quality for English–Spanish. We obtain consistent and statistically significant improvements in terms of BLEU and METEOR and we observe how the fused systems are able to handle synonyms to propose more adequate translations as well as help the system to disambiguate among several translation candidates for a word.

We present a simple new method where an emergent NMT system is used for simultaneously selecting training data and learning internal NMT representations. This is done in a self-supervised way without parallel data, in such a way that both tasks enhance each other during training. The method is language independent, introduces no additional hyper-parameters, and achieves BLEU scores of 29.21 (en2fr) and 27.36 (fr2en) on newstest2014 using English and French Wikipedia data for training.

This paper describes the UdS-DFKI participation to the multilingual task of the IWSLT Evaluation 2017. Our approach is based on factored multilingual neural translation systems following the small data and zero-shot training conditions. Our systems are designed to fully exploit multilinguality by including factors that increase the number of common elements among languages such as phonetic coarse encodings and synsets, besides shallow part-of-speech tags, stems and lemmas. Document level information is also considered by including the topic of every document. This approach improves a baseline without any additional factor for all the language pairs and even allows beyond-zero-shot translation. That is, the translation from unseen languages is possible thanks to the common elements —especially synsets in our models— among languages.

This is the Lump team participation at SemEval 2017 Task 1 on Semantic Textual Similarity. Our supervised model relies on features which are multilingual or interlingual in nature. We include lexical similarities, cross-language explicit semantic analysis, internal representations of multilingual neural networks and interlingual word embeddings. Our representations allow to use large datasets in language pairs with many instances to better classify instances in smaller language pairs avoiding the necessity of translating into a single language. Hence we can deal with all the languages in the task: Arabic, English, Spanish, and Turkish.

We propose a simple log-bilinear softmax-based model to deal with vocabulary expansion in machine translation. Our model uses word embeddings trained on significantly large unlabelled monolingual corpora and learns over a fairly small, word-to-word bilingual dictionary. Given an out-of-vocabulary source word, the model generates a probabilistic list of possible translations in the target language using the trained bilingual embeddings. We integrate these translation options into a standard phrase-based statistical machine translation system and obtain consistent improvements in translation quality on the English–Spanish language pair. When tested over an out-of-domain testset, we get a significant improvement of 3.9 BLEU points.

We introduce TweetMT, a parallel corpus of tweets in four language pairs that combine five languages (Spanish from/to Basque, Catalan, Galician and Portuguese), all of which have an official status in the Iberian Peninsula. The corpus has been created by combining automatic collection and crowdsourcing approaches, and it is publicly available. It is intended for the development and testing of microtext machine translation systems. In this paper we describe the methodology followed to build the corpus, and present the results of the shared task in which it was tested.

In this paper, we present a brief snapshot of the state of affairs in computational processing of Catalan and the initiatives that are starting to take place in an effort to bring the field a step forward, by making a better and more efficient use of the already existing resources and tools, by bridging the gap between research and market, and by establishing periodical meeting points for the community. In particular, we present the results of the First Workshop on the Computational Processing of Catalan, which succeeded in putting together a fair representation of the research in the area, and received attention from both the industry and the administration. Aside from facilitating communication among researchers and between developers and users, the Workshop provided the organizers with valuable information about existing resources, tools, developers and providers. This information has allowed us to go a step further by setting up a “harvesting” procedure which will hopefully build the seed of a portal-catalogue-observatory of language resources and technologies in Catalan.